Thread · 2 tweets · 19 Jun 2019

Replying to @matthewstoller and @LAM_Barrett, @daniellecitron, @benjaminwittes

So, a clear political bias would be different pricing in advertising for different constituencies. This could be in CPM of course, but also in higher CPOs. Both of these are dynamic. How do you audit that? How do you separate disparities of treatment vs impact?
How do you identify what is political? When Facebook tried to do that it was an unmitigated disaster. What happens when YouTube claims they aren't biased because they use machine learning?